Integrating Artificial Intelligence (AI) into Human Capital can enhance efficiency, improve decision-making, elevate employee experience, and provide valuable insights into workforce dynamics.
This presentation provides feedback on a Proof of Concept project that was undertaken in 2024 Q3 and the objective was to understand the various activities required for training an AI application on policy documents, refining cognitive search results using employee organizational data, and establishing and managing the Generative AI guardrails for the Human Capital domain.
As the organization is a financial services giant in Africa with a diverse technology and application landscape supporting multiple vendor solutions including SAP, the objective was to identify an AI application that is LLM-agnostic, allowing it to be configured to adopt the most suitable LLM for each AI use case. Another key requirement for the AI application is integration with key systems like SAP SuccessFactors. Due to the lack of in-house AI development capabilities, the decision was to implement an “off-the-shelf” generative AI and LLM application called Eerly Brain, developed by Accely, an SAP Partner.
Although there are many documented potential use cases the POC was based on the following three:
1. Conducting cognitive searches on a sample of selected company policy and FAQ documents while testing AI capabilities for indexing these documents at a specific folder location. This was to test setting the boundaries for the AI Agent.
2. Filtering cognitive search results based on the employee’s organizational data through the AI application’s integration with SAP SuccessFactors.
3. Exploring generative AI by setting guardrails specifically tailored for Human Capital, ensuring the AI application provides responses only for the human resources domain. For example, it will provide a response for “Suggest a job specification for a data analyst working in human capital” and not “How do I prune roses.”
Tangible Learnings
1. Implementing AI is a complex process. The AI application must be trained on the company’s structure, culture, acronyms, etc which involves setting appropriate guardrails.
2. Policy and FAQ documents need to be structured and, in some cases, rewritten to improve the accuracy of AI responses.
3. Maintaining a copy of LLMs on the organization’s cloud instance, regularly refreshing them, and setting guardrails can ensure that Generative AI applications are tailored to the organization and industry. This approach helps eliminate misuse and manage AI token usage costs.
